fs/nilfs2/super.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * super.c - NILFS module and super block management.
   4  *
   5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   6  *
   7  * Written by Ryusuke Konishi.
   8  */
   9 /*
  10  *  linux/fs/ext2/super.c
  11  *
  12  * Copyright (C) 1992, 1993, 1994, 1995
  13  * Remy Card ([email protected])
  14  * Laboratoire MASI - Institut Blaise Pascal
  15  * Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *  from
  18  *
  19  *  linux/fs/minix/inode.c
  20  *
  21  *  Copyright (C) 1991, 1992  Linus Torvalds
  22  *
  23  *  Big-endian to little-endian byte-swapping/bitmaps by
  24  *        David S. Miller ([email protected]), 1995
  25  */
  26
  27 #include <linux/module.h>
  28 #include <linux/string.h>
  29 #include <linux/slab.h>
  30 #include <linux/init.h>
  31 #include <linux/blkdev.h>
  32 #include <linux/parser.h>
  33 #include <linux/crc32.h>
  34 #include <linux/vfs.h>
  35 #include <linux/writeback.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/mount.h>
  38 #include "nilfs.h"
  39 #include "export.h"
  40 #include "mdt.h"
  41 #include "alloc.h"
  42 #include "btree.h"
  43 #include "btnode.h"
  44 #include "page.h"
  45 #include "cpfile.h"
  46 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
  47 #include "ifile.h"
  48 #include "dat.h"
  49 #include "segment.h"
  50 #include "segbuf.h"
  51
  52 MODULE_AUTHOR("NTT Corp.");
  53 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
  54                    "(NILFS)");
  55 MODULE_LICENSE("GPL");
  56
  57 static struct kmem_cache *nilfs_inode_cachep;
  58 struct kmem_cache *nilfs_transaction_cachep;
  59 struct kmem_cache *nilfs_segbuf_cachep;
  60 struct kmem_cache *nilfs_btree_path_cache;
  61
  62 static int nilfs_setup_super(struct super_block *sb, int is_mount);
  63 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
  64
  65 void __nilfs_msg(struct super_block *sb, const char *level, const char *fmt,
  66                  ...)
  67 {
  68         struct va_format vaf;
  69         va_list args;
  70
  71         va_start(args, fmt);
  72         vaf.fmt = fmt;
  73         vaf.va = &args;
  74         if (sb)
  75                 printk("%sNILFS (%s): %pV\n", level, sb->s_id, &vaf);
  76         else
  77                 printk("%sNILFS: %pV\n", level, &vaf);
  78         va_end(args);
  79 }
  80
  81 static void nilfs_set_error(struct super_block *sb)
  82 {
  83         struct the_nilfs *nilfs = sb->s_fs_info;
  84         struct nilfs_super_block **sbp;
  85
  86         down_write(&nilfs->ns_sem);
  87         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
  88                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
  89                 sbp = nilfs_prepare_super(sb, 0);
  90                 if (likely(sbp)) {
  91                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  92                         if (sbp[1])
  93                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
  94                         nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
  95                 }
  96         }
  97         up_write(&nilfs->ns_sem);
  98 }
  99
 100 /**
 101  * __nilfs_error() - report failure condition on a filesystem
 102  *
 103  * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
 104  * reporting an error message.  This function should be called when
 105  * NILFS detects incoherences or defects of meta data on disk.
 106  *
 107  * This implements the body of nilfs_error() macro.  Normally,
 108  * nilfs_error() should be used.  As for sustainable errors such as a
 109  * single-shot I/O error, nilfs_msg() should be used instead.
 110  *
 111  * Callers should not add a trailing newline since this will do it.
 112  */
 113 void __nilfs_error(struct super_block *sb, const char *function,
 114                    const char *fmt, ...)
 115 {
 116         struct the_nilfs *nilfs = sb->s_fs_info;
 117         struct va_format vaf;
 118         va_list args;
 119
 120         va_start(args, fmt);
 121
 122         vaf.fmt = fmt;
 123         vaf.va = &args;
 124
 125         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
 126                sb->s_id, function, &vaf);
 127
 128         va_end(args);
 129
 130         if (!sb_rdonly(sb)) {
 131                 nilfs_set_error(sb);
 132
 133                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
 134                         printk(KERN_CRIT "Remounting filesystem read-only\n");
 135                         sb->s_flags |= SB_RDONLY;
 136                 }
 137         }
 138
 139         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 140                 panic("NILFS (device %s): panic forced after error\n",
 141                       sb->s_id);
 142 }
 143
 144 struct inode *nilfs_alloc_inode(struct super_block *sb)
 145 {
 146         struct nilfs_inode_info *ii;
 147
 148         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
 149         if (!ii)
 150                 return NULL;
 151         ii->i_bh = NULL;
 152         ii->i_state = 0;
 153         ii->i_cno = 0;
 154         nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode);
 155         return &ii->vfs_inode;
 156 }
 157
 158 static void nilfs_i_callback(struct rcu_head *head)
 159 {
 160         struct inode *inode = container_of(head, struct inode, i_rcu);
 161
 162         if (nilfs_is_metadata_file_inode(inode))
 163                 nilfs_mdt_destroy(inode);
 164
 165         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
 166 }
 167
 168 void nilfs_destroy_inode(struct inode *inode)
 169 {
 170         call_rcu(&inode->i_rcu, nilfs_i_callback);
 171 }
 172
 173 static int nilfs_sync_super(struct super_block *sb, int flag)
 174 {
 175         struct the_nilfs *nilfs = sb->s_fs_info;
 176         int err;
 177
 178  retry:
 179         set_buffer_dirty(nilfs->ns_sbh[0]);
 180         if (nilfs_test_opt(nilfs, BARRIER)) {
 181                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
 182                                           REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
 183         } else {
 184                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
 185         }
 186
 187         if (unlikely(err)) {
 188                 nilfs_msg(sb, KERN_ERR, "unable to write superblock: err=%d",
 189                           err);
 190                 if (err == -EIO && nilfs->ns_sbh[1]) {
 191                         /*
 192                          * sbp[0] points to newer log than sbp[1],
 193                          * so copy sbp[0] to sbp[1] to take over sbp[0].
 194                          */
 195                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
 196                                nilfs->ns_sbsize);
 197                         nilfs_fall_back_super_block(nilfs);
 198                         goto retry;
 199                 }
 200         } else {
 201                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
 202
 203                 nilfs->ns_sbwcount++;
 204
 205                 /*
 206                  * The latest segment becomes trailable from the position
 207                  * written in superblock.
 208                  */
 209                 clear_nilfs_discontinued(nilfs);
 210
 211                 /* update GC protection for recent segments */
 212                 if (nilfs->ns_sbh[1]) {
 213                         if (flag == NILFS_SB_COMMIT_ALL) {
 214                                 set_buffer_dirty(nilfs->ns_sbh[1]);
 215                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
 216                                         goto out;
 217                         }
 218                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
 219                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
 220                                 sbp = nilfs->ns_sbp[1];
 221                 }
 222
 223                 spin_lock(&nilfs->ns_last_segment_lock);
 224                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
 225                 spin_unlock(&nilfs->ns_last_segment_lock);
 226         }
 227  out:
 228         return err;
 229 }
 230
 231 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
 232                           struct the_nilfs *nilfs)
 233 {
 234         sector_t nfreeblocks;
 235
 236         /* nilfs->ns_sem must be locked by the caller. */
 237         nilfs_count_free_blocks(nilfs, &nfreeblocks);
 238         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
 239
 240         spin_lock(&nilfs->ns_last_segment_lock);
 241         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
 242         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
 243         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
 244         spin_unlock(&nilfs->ns_last_segment_lock);
 245 }
 246
 247 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
 248                                                int flip)
 249 {
 250         struct the_nilfs *nilfs = sb->s_fs_info;
 251         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 252
 253         /* nilfs->ns_sem must be locked by the caller. */
 254         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 255                 if (sbp[1] &&
 256                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
 257                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
 258                 } else {
 259                         nilfs_msg(sb, KERN_CRIT, "superblock broke");
 260                         return NULL;
 261                 }
 262         } else if (sbp[1] &&
 263                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
 264                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 265         }
 266
 267         if (flip && sbp[1])
 268                 nilfs_swap_super_block(nilfs);
 269
 270         return sbp;
 271 }
 272
 273 int nilfs_commit_super(struct super_block *sb, int flag)
 274 {
 275         struct the_nilfs *nilfs = sb->s_fs_info;
 276         struct nilfs_super_block **sbp = nilfs->ns_sbp;
 277         time64_t t;
 278
 279         /* nilfs->ns_sem must be locked by the caller. */
 280         t = ktime_get_real_seconds();
 281         nilfs->ns_sbwtime = t;
 282         sbp[0]->s_wtime = cpu_to_le64(t);
 283         sbp[0]->s_sum = 0;
 284         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 285                                              (unsigned char *)sbp[0],
 286                                              nilfs->ns_sbsize));
 287         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
 288                 sbp[1]->s_wtime = sbp[0]->s_wtime;
 289                 sbp[1]->s_sum = 0;
 290                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
 291                                             (unsigned char *)sbp[1],
 292                                             nilfs->ns_sbsize));
 293         }
 294         clear_nilfs_sb_dirty(nilfs);
 295         nilfs->ns_flushed_device = 1;
 296         /* make sure store to ns_flushed_device cannot be reordered */
 297         smp_wmb();
 298         return nilfs_sync_super(sb, flag);
 299 }
 300
 301 /**
 302  * nilfs_cleanup_super() - write filesystem state for cleanup
 303  * @sb: super block instance to be unmounted or degraded to read-only
 304  *
 305  * This function restores state flags in the on-disk super block.
 306  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
 307  * filesystem was not clean previously.
 308  */
 309 int nilfs_cleanup_super(struct super_block *sb)
 310 {
 311         struct the_nilfs *nilfs = sb->s_fs_info;
 312         struct nilfs_super_block **sbp;
 313         int flag = NILFS_SB_COMMIT;
 314         int ret = -EIO;
 315
 316         sbp = nilfs_prepare_super(sb, 0);
 317         if (sbp) {
 318                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
 319                 nilfs_set_log_cursor(sbp[0], nilfs);
 320                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
 321                         /*
 322                          * make the "clean" flag also to the opposite
 323                          * super block if both super blocks point to
 324                          * the same checkpoint.
 325                          */
 326                         sbp[1]->s_state = sbp[0]->s_state;
 327                         flag = NILFS_SB_COMMIT_ALL;
 328                 }
 329                 ret = nilfs_commit_super(sb, flag);
 330         }
 331         return ret;
 332 }
 333
 334 /**
 335  * nilfs_move_2nd_super - relocate secondary super block
 336  * @sb: super block instance
 337  * @sb2off: new offset of the secondary super block (in bytes)
 338  */
 339 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
 340 {
 341         struct the_nilfs *nilfs = sb->s_fs_info;
 342         struct buffer_head *nsbh;
 343         struct nilfs_super_block *nsbp;
 344         sector_t blocknr, newblocknr;
 345         unsigned long offset;
 346         int sb2i;  /* array index of the secondary superblock */
 347         int ret = 0;
 348
 349         /* nilfs->ns_sem must be locked by the caller. */
 350         if (nilfs->ns_sbh[1] &&
 351             nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
 352                 sb2i = 1;
 353                 blocknr = nilfs->ns_sbh[1]->b_blocknr;
 354         } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
 355                 sb2i = 0;
 356                 blocknr = nilfs->ns_sbh[0]->b_blocknr;
 357         } else {
 358                 sb2i = -1;
 359                 blocknr = 0;
 360         }
 361         if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
 362                 goto out;  /* super block location is unchanged */
 363
 364         /* Get new super block buffer */
 365         newblocknr = sb2off >> nilfs->ns_blocksize_bits;
 366         offset = sb2off & (nilfs->ns_blocksize - 1);
 367         nsbh = sb_getblk(sb, newblocknr);
 368         if (!nsbh) {
 369                 nilfs_msg(sb, KERN_WARNING,
 370                           "unable to move secondary superblock to block %llu",
 371                           (unsigned long long)newblocknr);
 372                 ret = -EIO;
 373                 goto out;
 374         }
 375         nsbp = (void *)nsbh->b_data + offset;
 376         memset(nsbp, 0, nilfs->ns_blocksize);
 377
 378         if (sb2i >= 0) {
 379                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
 380                 brelse(nilfs->ns_sbh[sb2i]);
 381                 nilfs->ns_sbh[sb2i] = nsbh;
 382                 nilfs->ns_sbp[sb2i] = nsbp;
 383         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
 384                 /* secondary super block will be restored to index 1 */
 385                 nilfs->ns_sbh[1] = nsbh;
 386                 nilfs->ns_sbp[1] = nsbp;
 387         } else {
 388                 brelse(nsbh);
 389         }
 390 out:
 391         return ret;
 392 }
 393
 394 /**
 395  * nilfs_resize_fs - resize the filesystem
 396  * @sb: super block instance
 397  * @newsize: new size of the filesystem (in bytes)
 398  */
 399 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
 400 {
 401         struct the_nilfs *nilfs = sb->s_fs_info;
 402         struct nilfs_super_block **sbp;
 403         __u64 devsize, newnsegs;
 404         loff_t sb2off;
 405         int ret;
 406
 407         ret = -ERANGE;
 408         devsize = i_size_read(sb->s_bdev->bd_inode);
 409         if (newsize > devsize)
 410                 goto out;
 411
 412         /*
 413          * Write lock is required to protect some functions depending
 414          * on the number of segments, the number of reserved segments,
 415          * and so forth.
 416          */
 417         down_write(&nilfs->ns_segctor_sem);
 418
 419         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
 420         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
 421         do_div(newnsegs, nilfs->ns_blocks_per_segment);
 422
 423         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
 424         up_write(&nilfs->ns_segctor_sem);
 425         if (ret < 0)
 426                 goto out;
 427
 428         ret = nilfs_construct_segment(sb);
 429         if (ret < 0)
 430                 goto out;
 431
 432         down_write(&nilfs->ns_sem);
 433         nilfs_move_2nd_super(sb, sb2off);
 434         ret = -EIO;
 435         sbp = nilfs_prepare_super(sb, 0);
 436         if (likely(sbp)) {
 437                 nilfs_set_log_cursor(sbp[0], nilfs);
 438                 /*
 439                  * Drop NILFS_RESIZE_FS flag for compatibility with
 440                  * mount-time resize which may be implemented in a
 441                  * future release.
 442                  */
 443                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
 444                                               ~NILFS_RESIZE_FS);
 445                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
 446                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
 447                 if (sbp[1])
 448                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 449                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 450         }
 451         up_write(&nilfs->ns_sem);
 452
 453         /*
 454          * Reset the range of allocatable segments last.  This order
 455          * is important in the case of expansion because the secondary
 456          * superblock must be protected from log write until migration
 457          * completes.
 458          */
 459         if (!ret)
 460                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
 461 out:
 462         return ret;
 463 }
 464
 465 static void nilfs_put_super(struct super_block *sb)
 466 {
 467         struct the_nilfs *nilfs = sb->s_fs_info;
 468
 469         nilfs_detach_log_writer(sb);
 470
 471         if (!sb_rdonly(sb)) {
 472                 down_write(&nilfs->ns_sem);
 473                 nilfs_cleanup_super(sb);
 474                 up_write(&nilfs->ns_sem);
 475         }
 476
 477         iput(nilfs->ns_sufile);
 478         iput(nilfs->ns_cpfile);
 479         iput(nilfs->ns_dat);
 480
 481         destroy_nilfs(nilfs);
 482         sb->s_fs_info = NULL;
 483 }
 484
 485 static int nilfs_sync_fs(struct super_block *sb, int wait)
 486 {
 487         struct the_nilfs *nilfs = sb->s_fs_info;
 488         struct nilfs_super_block **sbp;
 489         int err = 0;
 490
 491         /* This function is called when super block should be written back */
 492         if (wait)
 493                 err = nilfs_construct_segment(sb);
 494
 495         down_write(&nilfs->ns_sem);
 496         if (nilfs_sb_dirty(nilfs)) {
 497                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
 498                 if (likely(sbp)) {
 499                         nilfs_set_log_cursor(sbp[0], nilfs);
 500                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
 501                 }
 502         }
 503         up_write(&nilfs->ns_sem);
 504
 505         if (!err)
 506                 err = nilfs_flush_device(nilfs);
 507
 508         return err;
 509 }
 510
 511 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
 512                             struct nilfs_root **rootp)
 513 {
 514         struct the_nilfs *nilfs = sb->s_fs_info;
 515         struct nilfs_root *root;
 516         struct nilfs_checkpoint *raw_cp;
 517         struct buffer_head *bh_cp;
 518         int err = -ENOMEM;
 519
 520         root = nilfs_find_or_create_root(
 521                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
 522         if (!root)
 523                 return err;
 524
 525         if (root->ifile)
 526                 goto reuse; /* already attached checkpoint */
 527
 528         down_read(&nilfs->ns_segctor_sem);
 529         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
 530                                           &bh_cp);
 531         up_read(&nilfs->ns_segctor_sem);
 532         if (unlikely(err)) {
 533                 if (err == -ENOENT || err == -EINVAL) {
 534                         nilfs_msg(sb, KERN_ERR,
 535                                   "Invalid checkpoint (checkpoint number=%llu)",
 536                                   (unsigned long long)cno);
 537                         err = -EINVAL;
 538                 }
 539                 goto failed;
 540         }
 541
 542         err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
 543                                &raw_cp->cp_ifile_inode, &root->ifile);
 544         if (err)
 545                 goto failed_bh;
 546
 547         atomic64_set(&root->inodes_count,
 548                         le64_to_cpu(raw_cp->cp_inodes_count));
 549         atomic64_set(&root->blocks_count,
 550                         le64_to_cpu(raw_cp->cp_blocks_count));
 551
 552         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 553
 554  reuse:
 555         *rootp = root;
 556         return 0;
 557
 558  failed_bh:
 559         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
 560  failed:
 561         nilfs_put_root(root);
 562
 563         return err;
 564 }
 565
 566 static int nilfs_freeze(struct super_block *sb)
 567 {
 568         struct the_nilfs *nilfs = sb->s_fs_info;
 569         int err;
 570
 571         if (sb_rdonly(sb))
 572                 return 0;
 573
 574         /* Mark super block clean */
 575         down_write(&nilfs->ns_sem);
 576         err = nilfs_cleanup_super(sb);
 577         up_write(&nilfs->ns_sem);
 578         return err;
 579 }
 580
 581 static int nilfs_unfreeze(struct super_block *sb)
 582 {
 583         struct the_nilfs *nilfs = sb->s_fs_info;
 584
 585         if (sb_rdonly(sb))
 586                 return 0;
 587
 588         down_write(&nilfs->ns_sem);
 589         nilfs_setup_super(sb, false);
 590         up_write(&nilfs->ns_sem);
 591         return 0;
 592 }
 593
 594 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 595 {
 596         struct super_block *sb = dentry->d_sb;
 597         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 598         struct the_nilfs *nilfs = root->nilfs;
 599         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
 600         unsigned long long blocks;
 601         unsigned long overhead;
 602         unsigned long nrsvblocks;
 603         sector_t nfreeblocks;
 604         u64 nmaxinodes, nfreeinodes;
 605         int err;
 606
 607         /*
 608          * Compute all of the segment blocks
 609          *
 610          * The blocks before first segment and after last segment
 611          * are excluded.
 612          */
 613         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
 614                 - nilfs->ns_first_data_block;
 615         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
 616
 617         /*
 618          * Compute the overhead
 619          *
 620          * When distributing meta data blocks outside segment structure,
 621          * We must count them as the overhead.
 622          */
 623         overhead = 0;
 624
 625         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
 626         if (unlikely(err))
 627                 return err;
 628
 629         err = nilfs_ifile_count_free_inodes(root->ifile,
 630                                             &nmaxinodes, &nfreeinodes);
 631         if (unlikely(err)) {
 632                 nilfs_msg(sb, KERN_WARNING,
 633                           "failed to count free inodes: err=%d", err);
 634                 if (err == -ERANGE) {
 635                         /*
 636                          * If nilfs_palloc_count_max_entries() returns
 637                          * -ERANGE error code then we simply treat
 638                          * curent inodes count as maximum possible and
 639                          * zero as free inodes value.
 640                          */
 641                         nmaxinodes = atomic64_read(&root->inodes_count);
 642                         nfreeinodes = 0;
 643                         err = 0;
 644                 } else
 645                         return err;
 646         }
 647
 648         buf->f_type = NILFS_SUPER_MAGIC;
 649         buf->f_bsize = sb->s_blocksize;
 650         buf->f_blocks = blocks - overhead;
 651         buf->f_bfree = nfreeblocks;
 652         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
 653                 (buf->f_bfree - nrsvblocks) : 0;
 654         buf->f_files = nmaxinodes;
 655         buf->f_ffree = nfreeinodes;
 656         buf->f_namelen = NILFS_NAME_LEN;
 657         buf->f_fsid.val[0] = (u32)id;
 658         buf->f_fsid.val[1] = (u32)(id >> 32);
 659
 660         return 0;
 661 }
 662
 663 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
 664 {
 665         struct super_block *sb = dentry->d_sb;
 666         struct the_nilfs *nilfs = sb->s_fs_info;
 667         struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
 668
 669         if (!nilfs_test_opt(nilfs, BARRIER))
 670                 seq_puts(seq, ",nobarrier");
 671         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
 672                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
 673         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
 674                 seq_puts(seq, ",errors=panic");
 675         if (nilfs_test_opt(nilfs, ERRORS_CONT))
 676                 seq_puts(seq, ",errors=continue");
 677         if (nilfs_test_opt(nilfs, STRICT_ORDER))
 678                 seq_puts(seq, ",order=strict");
 679         if (nilfs_test_opt(nilfs, NORECOVERY))
 680                 seq_puts(seq, ",norecovery");
 681         if (nilfs_test_opt(nilfs, DISCARD))
 682                 seq_puts(seq, ",discard");
 683
 684         return 0;
 685 }
 686
 687 static const struct super_operations nilfs_sops = {
 688         .alloc_inode    = nilfs_alloc_inode,
 689         .destroy_inode  = nilfs_destroy_inode,
 690         .dirty_inode    = nilfs_dirty_inode,
 691         .evict_inode    = nilfs_evict_inode,
 692         .put_super      = nilfs_put_super,
 693         .sync_fs        = nilfs_sync_fs,
 694         .freeze_fs      = nilfs_freeze,
 695         .unfreeze_fs    = nilfs_unfreeze,
 696         .statfs         = nilfs_statfs,
 697         .remount_fs     = nilfs_remount,
 698         .show_options = nilfs_show_options
 699 };
 700
 701 enum {
 702         Opt_err_cont, Opt_err_panic, Opt_err_ro,
 703         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
 704         Opt_discard, Opt_nodiscard, Opt_err,
 705 };
 706
 707 static match_table_t tokens = {
 708         {Opt_err_cont, "errors=continue"},
 709         {Opt_err_panic, "errors=panic"},
 710         {Opt_err_ro, "errors=remount-ro"},
 711         {Opt_barrier, "barrier"},
 712         {Opt_nobarrier, "nobarrier"},
 713         {Opt_snapshot, "cp=%u"},
 714         {Opt_order, "order=%s"},
 715         {Opt_norecovery, "norecovery"},
 716         {Opt_discard, "discard"},
 717         {Opt_nodiscard, "nodiscard"},
 718         {Opt_err, NULL}
 719 };
 720
 721 static int parse_options(char *options, struct super_block *sb, int is_remount)
 722 {
 723         struct the_nilfs *nilfs = sb->s_fs_info;
 724         char *p;
 725         substring_t args[MAX_OPT_ARGS];
 726
 727         if (!options)
 728                 return 1;
 729
 730         while ((p = strsep(&options, ",")) != NULL) {
 731                 int token;
 732
 733                 if (!*p)
 734                         continue;
 735
 736                 token = match_token(p, tokens, args);
 737                 switch (token) {
 738                 case Opt_barrier:
 739                         nilfs_set_opt(nilfs, BARRIER);
 740                         break;
 741                 case Opt_nobarrier:
 742                         nilfs_clear_opt(nilfs, BARRIER);
 743                         break;
 744                 case Opt_order:
 745                         if (strcmp(args[0].from, "relaxed") == 0)
 746                                 /* Ordered data semantics */
 747                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
 748                         else if (strcmp(args[0].from, "strict") == 0)
 749                                 /* Strict in-order semantics */
 750                                 nilfs_set_opt(nilfs, STRICT_ORDER);
 751                         else
 752                                 return 0;
 753                         break;
 754                 case Opt_err_panic:
 755                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
 756                         break;
 757                 case Opt_err_ro:
 758                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
 759                         break;
 760                 case Opt_err_cont:
 761                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
 762                         break;
 763                 case Opt_snapshot:
 764                         if (is_remount) {
 765                                 nilfs_msg(sb, KERN_ERR,
 766                                           "\"%s\" option is invalid for remount",
 767                                           p);
 768                                 return 0;
 769                         }
 770                         break;
 771                 case Opt_norecovery:
 772                         nilfs_set_opt(nilfs, NORECOVERY);
 773                         break;
 774                 case Opt_discard:
 775                         nilfs_set_opt(nilfs, DISCARD);
 776                         break;
 777                 case Opt_nodiscard:
 778                         nilfs_clear_opt(nilfs, DISCARD);
 779                         break;
 780                 default:
 781                         nilfs_msg(sb, KERN_ERR,
 782                                   "unrecognized mount option \"%s\"", p);
 783                         return 0;
 784                 }
 785         }
 786         return 1;
 787 }
 788
 789 static inline void
 790 nilfs_set_default_options(struct super_block *sb,
 791                           struct nilfs_super_block *sbp)
 792 {
 793         struct the_nilfs *nilfs = sb->s_fs_info;
 794
 795         nilfs->ns_mount_opt =
 796                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
 797 }
 798
 799 static int nilfs_setup_super(struct super_block *sb, int is_mount)
 800 {
 801         struct the_nilfs *nilfs = sb->s_fs_info;
 802         struct nilfs_super_block **sbp;
 803         int max_mnt_count;
 804         int mnt_count;
 805
 806         /* nilfs->ns_sem must be locked by the caller. */
 807         sbp = nilfs_prepare_super(sb, 0);
 808         if (!sbp)
 809                 return -EIO;
 810
 811         if (!is_mount)
 812                 goto skip_mount_setup;
 813
 814         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
 815         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
 816
 817         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
 818                 nilfs_msg(sb, KERN_WARNING, "mounting fs with errors");
 819 #if 0
 820         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
 821                 nilfs_msg(sb, KERN_WARNING, "maximal mount count reached");
 822 #endif
 823         }
 824         if (!max_mnt_count)
 825                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
 826
 827         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
 828         sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
 829
 830 skip_mount_setup:
 831         sbp[0]->s_state =
 832                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
 833         /* synchronize sbp[1] with sbp[0] */
 834         if (sbp[1])
 835                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
 836         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
 837 }
 838
 839 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
 840                                                  u64 pos, int blocksize,
 841                                                  struct buffer_head **pbh)
 842 {
 843         unsigned long long sb_index = pos;
 844         unsigned long offset;
 845
 846         offset = do_div(sb_index, blocksize);
 847         *pbh = sb_bread(sb, sb_index);
 848         if (!*pbh)
 849                 return NULL;
 850         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
 851 }
 852
 853 int nilfs_store_magic_and_option(struct super_block *sb,
 854                                  struct nilfs_super_block *sbp,
 855                                  char *data)
 856 {
 857         struct the_nilfs *nilfs = sb->s_fs_info;
 858
 859         sb->s_magic = le16_to_cpu(sbp->s_magic);
 860
 861         /* FS independent flags */
 862 #ifdef NILFS_ATIME_DISABLE
 863         sb->s_flags |= SB_NOATIME;
 864 #endif
 865
 866         nilfs_set_default_options(sb, sbp);
 867
 868         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
 869         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
 870         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
 871         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
 872
 873         return !parse_options(data, sb, 0) ? -EINVAL : 0;
 874 }
 875
 876 int nilfs_check_feature_compatibility(struct super_block *sb,
 877                                       struct nilfs_super_block *sbp)
 878 {
 879         __u64 features;
 880
 881         features = le64_to_cpu(sbp->s_feature_incompat) &
 882                 ~NILFS_FEATURE_INCOMPAT_SUPP;
 883         if (features) {
 884                 nilfs_msg(sb, KERN_ERR,
 885                           "couldn't mount because of unsupported optional features (%llx)",
 886                           (unsigned long long)features);
 887                 return -EINVAL;
 888         }
 889         features = le64_to_cpu(sbp->s_feature_compat_ro) &
 890                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
 891         if (!sb_rdonly(sb) && features) {
 892                 nilfs_msg(sb, KERN_ERR,
 893                           "couldn't mount RDWR because of unsupported optional features (%llx)",
 894                           (unsigned long long)features);
 895                 return -EINVAL;
 896         }
 897         return 0;
 898 }
 899
 900 static int nilfs_get_root_dentry(struct super_block *sb,
 901                                  struct nilfs_root *root,
 902                                  struct dentry **root_dentry)
 903 {
 904         struct inode *inode;
 905         struct dentry *dentry;
 906         int ret = 0;
 907
 908         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
 909         if (IS_ERR(inode)) {
 910                 ret = PTR_ERR(inode);
 911                 nilfs_msg(sb, KERN_ERR, "error %d getting root inode", ret);
 912                 goto out;
 913         }
 914         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
 915                 iput(inode);
 916                 nilfs_msg(sb, KERN_ERR, "corrupt root inode");
 917                 ret = -EINVAL;
 918                 goto out;
 919         }
 920
 921         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
 922                 dentry = d_find_alias(inode);
 923                 if (!dentry) {
 924                         dentry = d_make_root(inode);
 925                         if (!dentry) {
 926                                 ret = -ENOMEM;
 927                                 goto failed_dentry;
 928                         }
 929                 } else {
 930                         iput(inode);
 931                 }
 932         } else {
 933                 dentry = d_obtain_root(inode);
 934                 if (IS_ERR(dentry)) {
 935                         ret = PTR_ERR(dentry);
 936                         goto failed_dentry;
 937                 }
 938         }
 939         *root_dentry = dentry;
 940  out:
 941         return ret;
 942
 943  failed_dentry:
 944         nilfs_msg(sb, KERN_ERR, "error %d getting root dentry", ret);
 945         goto out;
 946 }
 947
 948 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
 949                                  struct dentry **root_dentry)
 950 {
 951         struct the_nilfs *nilfs = s->s_fs_info;
 952         struct nilfs_root *root;
 953         int ret;
 954
 955         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
 956
 957         down_read(&nilfs->ns_segctor_sem);
 958         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
 959         up_read(&nilfs->ns_segctor_sem);
 960         if (ret < 0) {
 961                 ret = (ret == -ENOENT) ? -EINVAL : ret;
 962                 goto out;
 963         } else if (!ret) {
 964                 nilfs_msg(s, KERN_ERR,
 965                           "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
 966                           (unsigned long long)cno);
 967                 ret = -EINVAL;
 968                 goto out;
 969         }
 970
 971         ret = nilfs_attach_checkpoint(s, cno, false, &root);
 972         if (ret) {
 973                 nilfs_msg(s, KERN_ERR,
 974                           "error %d while loading snapshot (checkpoint number=%llu)",
 975                           ret, (unsigned long long)cno);
 976                 goto out;
 977         }
 978         ret = nilfs_get_root_dentry(s, root, root_dentry);
 979         nilfs_put_root(root);
 980  out:
 981         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
 982         return ret;
 983 }
 984
 985 /**
 986  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
 987  * @root_dentry: root dentry of the tree to be shrunk
 988  *
 989  * This function returns true if the tree was in-use.
 990  */
 991 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
 992 {
 993         shrink_dcache_parent(root_dentry);
 994         return d_count(root_dentry) > 1;
 995 }
 996
 997 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
 998 {
 999         struct the_nilfs *nilfs = sb->s_fs_info;
1000         struct nilfs_root *root;
1001         struct inode *inode;
1002         struct dentry *dentry;
1003         int ret;
1004
1005         if (cno > nilfs->ns_cno)
1006                 return false;
1007
1008         if (cno >= nilfs_last_cno(nilfs))
1009                 return true;    /* protect recent checkpoints */
1010
1011         ret = false;
1012         root = nilfs_lookup_root(nilfs, cno);
1013         if (root) {
1014                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1015                 if (inode) {
1016                         dentry = d_find_alias(inode);
1017                         if (dentry) {
1018                                 ret = nilfs_tree_is_busy(dentry);
1019                                 dput(dentry);
1020                         }
1021                         iput(inode);
1022                 }
1023                 nilfs_put_root(root);
1024         }
1025         return ret;
1026 }
1027
1028 /**
1029  * nilfs_fill_super() - initialize a super block instance
1030  * @sb: super_block
1031  * @data: mount options
1032  * @silent: silent mode flag
1033  *
1034  * This function is called exclusively by nilfs->ns_mount_mutex.
1035  * So, the recovery process is protected from other simultaneous mounts.
1036  */
1037 static int
1038 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1039 {
1040         struct the_nilfs *nilfs;
1041         struct nilfs_root *fsroot;
1042         __u64 cno;
1043         int err;
1044
1045         nilfs = alloc_nilfs(sb);
1046         if (!nilfs)
1047                 return -ENOMEM;
1048
1049         sb->s_fs_info = nilfs;
1050
1051         err = init_nilfs(nilfs, sb, (char *)data);
1052         if (err)
1053                 goto failed_nilfs;
1054
1055         sb->s_op = &nilfs_sops;
1056         sb->s_export_op = &nilfs_export_ops;
1057         sb->s_root = NULL;
1058         sb->s_time_gran = 1;
1059         sb->s_max_links = NILFS_LINK_MAX;
1060
1061         sb->s_bdi = bdi_get(sb->s_bdev->bd_bdi);
1062
1063         err = load_nilfs(nilfs, sb);
1064         if (err)
1065                 goto failed_nilfs;
1066
1067         cno = nilfs_last_cno(nilfs);
1068         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1069         if (err) {
1070                 nilfs_msg(sb, KERN_ERR,
1071                           "error %d while loading last checkpoint (checkpoint number=%llu)",
1072                           err, (unsigned long long)cno);
1073                 goto failed_unload;
1074         }
1075
1076         if (!sb_rdonly(sb)) {
1077                 err = nilfs_attach_log_writer(sb, fsroot);
1078                 if (err)
1079                         goto failed_checkpoint;
1080         }
1081
1082         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1083         if (err)
1084                 goto failed_segctor;
1085
1086         nilfs_put_root(fsroot);
1087
1088         if (!sb_rdonly(sb)) {
1089                 down_write(&nilfs->ns_sem);
1090                 nilfs_setup_super(sb, true);
1091                 up_write(&nilfs->ns_sem);
1092         }
1093
1094         return 0;
1095
1096  failed_segctor:
1097         nilfs_detach_log_writer(sb);
1098
1099  failed_checkpoint:
1100         nilfs_put_root(fsroot);
1101
1102  failed_unload:
1103         iput(nilfs->ns_sufile);
1104         iput(nilfs->ns_cpfile);
1105         iput(nilfs->ns_dat);
1106
1107  failed_nilfs:
1108         destroy_nilfs(nilfs);
1109         return err;
1110 }
1111
1112 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1113 {
1114         struct the_nilfs *nilfs = sb->s_fs_info;
1115         unsigned long old_sb_flags;
1116         unsigned long old_mount_opt;
1117         int err;
1118
1119         sync_filesystem(sb);
1120         old_sb_flags = sb->s_flags;
1121         old_mount_opt = nilfs->ns_mount_opt;
1122
1123         if (!parse_options(data, sb, 1)) {
1124                 err = -EINVAL;
1125                 goto restore_opts;
1126         }
1127         sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1128
1129         err = -EINVAL;
1130
1131         if (!nilfs_valid_fs(nilfs)) {
1132                 nilfs_msg(sb, KERN_WARNING,
1133                           "couldn't remount because the filesystem is in an incomplete recovery state");
1134                 goto restore_opts;
1135         }
1136
1137         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1138                 goto out;
1139         if (*flags & SB_RDONLY) {
1140                 /* Shutting down log writer */
1141                 nilfs_detach_log_writer(sb);
1142                 sb->s_flags |= SB_RDONLY;
1143
1144                 /*
1145                  * Remounting a valid RW partition RDONLY, so set
1146                  * the RDONLY flag and then mark the partition as valid again.
1147                  */
1148                 down_write(&nilfs->ns_sem);
1149                 nilfs_cleanup_super(sb);
1150                 up_write(&nilfs->ns_sem);
1151         } else {
1152                 __u64 features;
1153                 struct nilfs_root *root;
1154
1155                 /*
1156                  * Mounting a RDONLY partition read-write, so reread and
1157                  * store the current valid flag.  (It may have been changed
1158                  * by fsck since we originally mounted the partition.)
1159                  */
1160                 down_read(&nilfs->ns_sem);
1161                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1162                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1163                 up_read(&nilfs->ns_sem);
1164                 if (features) {
1165                         nilfs_msg(sb, KERN_WARNING,
1166                                   "couldn't remount RDWR because of unsupported optional features (%llx)",
1167                                   (unsigned long long)features);
1168                         err = -EROFS;
1169                         goto restore_opts;
1170                 }
1171
1172                 sb->s_flags &= ~SB_RDONLY;
1173
1174                 root = NILFS_I(d_inode(sb->s_root))->i_root;
1175                 err = nilfs_attach_log_writer(sb, root);
1176                 if (err)
1177                         goto restore_opts;
1178
1179                 down_write(&nilfs->ns_sem);
1180                 nilfs_setup_super(sb, true);
1181                 up_write(&nilfs->ns_sem);
1182         }
1183  out:
1184         return 0;
1185
1186  restore_opts:
1187         sb->s_flags = old_sb_flags;
1188         nilfs->ns_mount_opt = old_mount_opt;
1189         return err;
1190 }
1191
1192 struct nilfs_super_data {
1193         struct block_device *bdev;
1194         __u64 cno;
1195         int flags;
1196 };
1197
1198 static int nilfs_parse_snapshot_option(const char *option,
1199                                        const substring_t *arg,
1200                                        struct nilfs_super_data *sd)
1201 {
1202         unsigned long long val;
1203         const char *msg = NULL;
1204         int err;
1205
1206         if (!(sd->flags & SB_RDONLY)) {
1207                 msg = "read-only option is not specified";
1208                 goto parse_error;
1209         }
1210
1211         err = kstrtoull(arg->from, 0, &val);
1212         if (err) {
1213                 if (err == -ERANGE)
1214                         msg = "too large checkpoint number";
1215                 else
1216                         msg = "malformed argument";
1217                 goto parse_error;
1218         } else if (val == 0) {
1219                 msg = "invalid checkpoint number 0";
1220                 goto parse_error;
1221         }
1222         sd->cno = val;
1223         return 0;
1224
1225 parse_error:
1226         nilfs_msg(NULL, KERN_ERR, "invalid option \"%s\": %s", option, msg);
1227         return 1;
1228 }
1229
1230 /**
1231  * nilfs_identify - pre-read mount options needed to identify mount instance
1232  * @data: mount options
1233  * @sd: nilfs_super_data
1234  */
1235 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1236 {
1237         char *p, *options = data;
1238         substring_t args[MAX_OPT_ARGS];
1239         int token;
1240         int ret = 0;
1241
1242         do {
1243                 p = strsep(&options, ",");
1244                 if (p != NULL && *p) {
1245                         token = match_token(p, tokens, args);
1246                         if (token == Opt_snapshot)
1247                                 ret = nilfs_parse_snapshot_option(p, &args[0],
1248                                                                   sd);
1249                 }
1250                 if (!options)
1251                         break;
1252                 BUG_ON(options == data);
1253                 *(options - 1) = ',';
1254         } while (!ret);
1255         return ret;
1256 }
1257
1258 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1259 {
1260         s->s_bdev = data;
1261         s->s_dev = s->s_bdev->bd_dev;
1262         return 0;
1263 }
1264
1265 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1266 {
1267         return (void *)s->s_bdev == data;
1268 }
1269
1270 static struct dentry *
1271 nilfs_mount(struct file_system_type *fs_type, int flags,
1272              const char *dev_name, void *data)
1273 {
1274         struct nilfs_super_data sd;
1275         struct super_block *s;
1276         fmode_t mode = FMODE_READ | FMODE_EXCL;
1277         struct dentry *root_dentry;
1278         int err, s_new = false;
1279
1280         if (!(flags & SB_RDONLY))
1281                 mode |= FMODE_WRITE;
1282
1283         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1284         if (IS_ERR(sd.bdev))
1285                 return ERR_CAST(sd.bdev);
1286
1287         sd.cno = 0;
1288         sd.flags = flags;
1289         if (nilfs_identify((char *)data, &sd)) {
1290                 err = -EINVAL;
1291                 goto failed;
1292         }
1293
1294         /*
1295          * once the super is inserted into the list by sget, s_umount
1296          * will protect the lockfs code from trying to start a snapshot
1297          * while we are mounting
1298          */
1299         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1300         if (sd.bdev->bd_fsfreeze_count > 0) {
1301                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1302                 err = -EBUSY;
1303                 goto failed;
1304         }
1305         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1306                  sd.bdev);
1307         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1308         if (IS_ERR(s)) {
1309                 err = PTR_ERR(s);
1310                 goto failed;
1311         }
1312
1313         if (!s->s_root) {
1314                 s_new = true;
1315
1316                 /* New superblock instance created */
1317                 s->s_mode = mode;
1318                 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1319                 sb_set_blocksize(s, block_size(sd.bdev));
1320
1321                 err = nilfs_fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1322                 if (err)
1323                         goto failed_super;
1324
1325                 s->s_flags |= SB_ACTIVE;
1326         } else if (!sd.cno) {
1327                 if (nilfs_tree_is_busy(s->s_root)) {
1328                         if ((flags ^ s->s_flags) & SB_RDONLY) {
1329                                 nilfs_msg(s, KERN_ERR,
1330                                           "the device already has a %s mount.",
1331                                           sb_rdonly(s) ? "read-only" : "read/write");
1332                                 err = -EBUSY;
1333                                 goto failed_super;
1334                         }
1335                 } else {
1336                         /*
1337                          * Try remount to setup mount states if the current
1338                          * tree is not mounted and only snapshots use this sb.
1339                          */
1340                         err = nilfs_remount(s, &flags, data);
1341                         if (err)
1342                                 goto failed_super;
1343                 }
1344         }
1345
1346         if (sd.cno) {
1347                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1348                 if (err)
1349                         goto failed_super;
1350         } else {
1351                 root_dentry = dget(s->s_root);
1352         }
1353
1354         if (!s_new)
1355                 blkdev_put(sd.bdev, mode);
1356
1357         return root_dentry;
1358
1359  failed_super:
1360         deactivate_locked_super(s);
1361
1362  failed:
1363         if (!s_new)
1364                 blkdev_put(sd.bdev, mode);
1365         return ERR_PTR(err);
1366 }
1367
1368 struct file_system_type nilfs_fs_type = {
1369         .owner    = THIS_MODULE,
1370         .name     = "nilfs2",
1371         .mount    = nilfs_mount,
1372         .kill_sb  = kill_block_super,
1373         .fs_flags = FS_REQUIRES_DEV,
1374 };
1375 MODULE_ALIAS_FS("nilfs2");
1376
1377 static void nilfs_inode_init_once(void *obj)
1378 {
1379         struct nilfs_inode_info *ii = obj;
1380
1381         INIT_LIST_HEAD(&ii->i_dirty);
1382 #ifdef CONFIG_NILFS_XATTR
1383         init_rwsem(&ii->xattr_sem);
1384 #endif
1385         address_space_init_once(&ii->i_btnode_cache);
1386         ii->i_bmap = &ii->i_bmap_data;
1387         inode_init_once(&ii->vfs_inode);
1388 }
1389
1390 static void nilfs_segbuf_init_once(void *obj)
1391 {
1392         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1393 }
1394
1395 static void nilfs_destroy_cachep(void)
1396 {
1397         /*
1398          * Make sure all delayed rcu free inodes are flushed before we
1399          * destroy cache.
1400          */
1401         rcu_barrier();
1402
1403         kmem_cache_destroy(nilfs_inode_cachep);
1404         kmem_cache_destroy(nilfs_transaction_cachep);
1405         kmem_cache_destroy(nilfs_segbuf_cachep);
1406         kmem_cache_destroy(nilfs_btree_path_cache);
1407 }
1408
1409 static int __init nilfs_init_cachep(void)
1410 {
1411         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1412                         sizeof(struct nilfs_inode_info), 0,
1413                         SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1414                         nilfs_inode_init_once);
1415         if (!nilfs_inode_cachep)
1416                 goto fail;
1417
1418         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1419                         sizeof(struct nilfs_transaction_info), 0,
1420                         SLAB_RECLAIM_ACCOUNT, NULL);
1421         if (!nilfs_transaction_cachep)
1422                 goto fail;
1423
1424         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1425                         sizeof(struct nilfs_segment_buffer), 0,
1426                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1427         if (!nilfs_segbuf_cachep)
1428                 goto fail;
1429
1430         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1431                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1432                         0, 0, NULL);
1433         if (!nilfs_btree_path_cache)
1434                 goto fail;
1435
1436         return 0;
1437
1438 fail:
1439         nilfs_destroy_cachep();
1440         return -ENOMEM;
1441 }
1442
1443 static int __init init_nilfs_fs(void)
1444 {
1445         int err;
1446
1447         err = nilfs_init_cachep();
1448         if (err)
1449                 goto fail;
1450
1451         err = nilfs_sysfs_init();
1452         if (err)
1453                 goto free_cachep;
1454
1455         err = register_filesystem(&nilfs_fs_type);
1456         if (err)
1457                 goto deinit_sysfs_entry;
1458
1459         printk(KERN_INFO "NILFS version 2 loaded\n");
1460         return 0;
1461
1462 deinit_sysfs_entry:
1463         nilfs_sysfs_exit();
1464 free_cachep:
1465         nilfs_destroy_cachep();
1466 fail:
1467         return err;
1468 }
1469
1470 static void __exit exit_nilfs_fs(void)
1471 {
1472         nilfs_destroy_cachep();
1473         nilfs_sysfs_exit();
1474         unregister_filesystem(&nilfs_fs_type);
1475 }
1476
1477 module_init(init_nilfs_fs)
1478 module_exit(exit_nilfs_fs)